Files
nebula/control.go
Matt RichardsonandClaude Opus 5 14a4d87faf Add nebula ctl, a local socket for the debug commands
Every diagnostic command nebula has was reachable through exactly one door:
the built-in ssh debug server. That server is off by default, and turning it
on means generating a host key, writing an sshd block with authorized public
keys, and SIGHUPing the daemon. That is a lot of ceremony to answer "what
version is this node running".

Nebula now serves the same commands over a local unix socket, enabled by
default, and `nebula ctl <command>` runs them. The socket lives in a 0700
directory so filesystem permissions are the access control; no keys, nothing
on the network. Failing to create it is logged and never blocks startup.

The command registry was already transport neutral, so this is mostly new
transport rather than new commands:

  - diag/ holds the registry, dispatch, writer and wire protocol, moved out
    of sshd because none of it was ever about ssh. sshd and ctl.go dispatch
    against one shared registry.
  - commands.go holds every command implementation, moved out of ssh.go
    (which was 85% not ssh) and renamed off the ssh prefix. Adding a command
    there makes it available over both transports.
  - ssh.go keeps only host keys, authorized users, and the listen address.
  - ctl.go supervises the socket, following the statsServer lifecycle shape.

The wire protocol frames the response rather than terminating it, because
print-cert -raw and list-hostmap -json both emit arbitrary bytes that no
sentinel could safely delimit. argv travels as a list so quoting survives.
Exit statuses are real: 0, 2 for usage, 127 for an unknown command.

Two things fall out. The ssh console now reports a real exit status instead
of a hardcoded zero, so `ssh host list-hostmap` is scriptable too. And eight
command callbacks that silently returned nil on a flags type mismatch now
report it, which the exit status makes visible.

Windows is a stub returning a clear "not supported" until it gets a named
pipe with a security descriptor; iOS and Android are never enabled, having no
daemon for a CLI to attach to.

Breaking for embedders of the sshd package: NewSSHServer takes a
*diag.Registry, SSHServer.RegisterCommand is gone in favor of registering on
that registry, and the command types live in diag rather than sshd.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014fya5fTXGiwX72FUmoL9y3
2026-09-09 17:07:35 -04:00

426 lines
12 KiB
Go

package nebula
import (
"context"
"errors"
"log/slog"
"net/netip"
"os"
"os/signal"
"sync"
"syscall"
"github.com/slackhq/nebula/cert"
"github.com/slackhq/nebula/header"
"github.com/slackhq/nebula/overlay"
)
type RunState int
const (
StateUnknown RunState = iota
StateReady
StateStarted
StateStopping
StateStopped
)
var ErrAlreadyStarted = errors.New("nebula is already started")
var ErrAlreadyStopped = errors.New("nebula cannot be restarted")
var ErrUnknownState = errors.New("nebula state is invalid")
// Every interaction here needs to take extra care to copy memory and not return or use arguments "as is" when touching
// core. This means copying IP objects, slices, de-referencing pointers and taking the actual value, etc
type controlEach func(h *HostInfo)
type controlHostLister interface {
QueryVpnAddr(vpnAddr netip.Addr) *HostInfo
ForEachIndex(each controlEach)
ForEachVpnAddr(each controlEach)
GetPreferredRanges() []netip.Prefix
}
type Control struct {
stateLock sync.Mutex
state RunState
f *Interface
l *slog.Logger
ctx context.Context
cancel context.CancelFunc
sshStart func()
ctlStart func()
statsStart func()
dnsStart func()
lighthouseStart func()
networkChangeStart func(rebind func())
connectionManagerStart func(context.Context)
}
type ControlHostInfo struct {
VpnAddrs []netip.Addr `json:"vpnAddrs"`
LocalIndex uint32 `json:"localIndex"`
RemoteIndex uint32 `json:"remoteIndex"`
RemoteAddrs []netip.AddrPort `json:"remoteAddrs"`
Cert cert.Certificate `json:"cert"`
MessageCounter uint64 `json:"messageCounter"`
CurrentRemote netip.AddrPort `json:"currentRemote"`
CurrentRelaysToMe []netip.Addr `json:"currentRelaysToMe"`
CurrentRelaysThroughMe []netip.Addr `json:"currentRelaysThroughMe"`
}
// Start actually runs nebula, this is a nonblocking call.
// Use Wait to block until nebula has fully stopped and to learn whether a fatal reader error caused the shutdown.
func (c *Control) Start() error {
c.stateLock.Lock()
defer c.stateLock.Unlock()
switch c.state {
case StateReady:
//yay!
case StateStopped, StateStopping:
return ErrAlreadyStopped
case StateStarted:
return ErrAlreadyStarted
default:
return ErrUnknownState
}
// Activate the interface
err := c.f.activate()
if err != nil {
// Cancel before Close so a caller returning from Wait always observes a dead Context
c.cancel()
_ = c.f.Close()
c.state = StateStopped
return err
}
// Call all the delayed funcs that waited patiently for the interface to be created.
if c.sshStart != nil {
go c.sshStart()
}
if c.ctlStart != nil {
go c.ctlStart()
}
if c.statsStart != nil {
go c.statsStart()
}
if c.dnsStart != nil {
go c.dnsStart()
}
if c.networkChangeStart != nil {
go c.networkChangeStart(c.RebindUDPServer)
}
if c.connectionManagerStart != nil {
go c.connectionManagerStart(c.ctx)
}
if c.lighthouseStart != nil {
c.lighthouseStart()
}
c.f.triggerShutdown = func() { go c.Stop() }
// Start reading packets.
c.f.run()
c.state = StateStarted
return nil
}
func (c *Control) State() RunState {
c.stateLock.Lock()
defer c.stateLock.Unlock()
return c.state
}
func (c *Control) Context() context.Context {
return c.ctx
}
// Stop tears nebula down, closing all tunnels and releasing everything it holds.
// Use Wait to block until the shutdown has completed.
// A Control that has been stopped cannot be started again, Start will return ErrAlreadyStopped.
func (c *Control) Stop() {
c.stateLock.Lock()
switch c.state {
case StateStarted:
// Fall through to the full teardown below
case StateReady:
// Never started
c.cancel()
c.state = StateStopped
if err := c.f.Close(); err != nil {
c.l.Error("Close interface failed", "error", err)
}
c.stateLock.Unlock()
return
default:
c.stateLock.Unlock()
// We are stopping or stopped already
return
}
c.state = StateStopping
c.stateLock.Unlock()
// Closing tunnels can be slow with a large hostmap, don't hold the lock for it
c.cancel()
c.CloseAllTunnels(false)
c.stateLock.Lock()
c.state = StateStopped
if err := c.f.Close(); err != nil {
c.l.Error("Close interface failed", "error", err)
}
c.stateLock.Unlock()
}
// Wait blocks until nebula has fully stopped, either via Stop or an internal fatal error,
// and returns the first fatal packet reader error if there was one.
// It is safe to call from multiple goroutines and at any point in the lifecycle,
// but a Wait on a Control that is never started and never stopped will block forever.
func (c *Control) Wait() error {
return c.f.wait()
}
// ShutdownBlock will listen for and block on term and interrupt signals, calling Control.Stop() once signalled
func (c *Control) ShutdownBlock() {
sigChan := make(chan os.Signal, 1)
signal.Notify(sigChan, syscall.SIGTERM)
signal.Notify(sigChan, syscall.SIGINT)
rawSig := <-sigChan
sig := rawSig.String()
c.l.Info("Caught signal, shutting down", "signal", sig)
c.Stop()
}
// RebindUDPServer asks the UDP listener to rebind it's listener. Mainly used on mobile clients when interfaces change.
func (c *Control) RebindUDPServer() {
c.stateLock.Lock()
defer c.stateLock.Unlock()
if c.state != StateStarted {
return
}
// A failure here means we are likely still pinned to the interface we came up on, so the rest of this is
// unlikely to help. Say so instead of silently carrying on as if we rebound.
if err := c.f.outside.Rebind(); err != nil {
c.l.Error("Failed to rebind udp socket", "error", err)
}
// Trigger a lighthouse update, useful for mobile clients that should have an update interval of 0
c.f.lightHouse.SendUpdate()
// Let the main interface know that we rebound so that underlying tunnels know to trigger punches from their remotes
c.f.rebindEpoch.Add(1)
}
// ListHostmapHosts returns details about the actual or pending (handshaking) hostmap by vpn ip
func (c *Control) ListHostmapHosts(pendingMap bool) []ControlHostInfo {
if pendingMap {
return listHostMapHosts(c.f.handshakeManager)
} else {
return listHostMapHosts(c.f.hostMap)
}
}
// ListHostmapIndexes returns details about the actual or pending (handshaking) hostmap by local index id
func (c *Control) ListHostmapIndexes(pendingMap bool) []ControlHostInfo {
if pendingMap {
return listHostMapIndexes(c.f.handshakeManager)
} else {
return listHostMapIndexes(c.f.hostMap)
}
}
// GetCertByVpnIp returns the authenticated certificate of the given vpn IP, or nil if not found
func (c *Control) GetCertByVpnIp(vpnIp netip.Addr) cert.Certificate {
if c.f.myVpnAddrsTable.Contains(vpnIp) {
// Only returning the default certificate since its impossible
// for any other host but ourselves to have more than 1
return c.f.pki.getCertState().GetDefaultCertificate().Copy()
}
hi := c.f.hostMap.QueryVpnAddr(vpnIp)
if hi == nil {
return nil
}
return hi.GetCert().Certificate.Copy()
}
// CreateTunnel creates a new tunnel to the given vpn ip.
func (c *Control) CreateTunnel(vpnIp netip.Addr) {
c.f.handshakeManager.StartHandshake(vpnIp, nil)
}
// PrintTunnel creates a new tunnel to the given vpn ip.
func (c *Control) PrintTunnel(vpnIp netip.Addr) *ControlHostInfo {
hi := c.f.hostMap.QueryVpnAddr(vpnIp)
if hi == nil {
return nil
}
chi := copyHostInfo(hi, c.f.hostMap.GetPreferredRanges())
return &chi
}
// QueryLighthouse queries the lighthouse.
func (c *Control) QueryLighthouse(vpnIp netip.Addr) *CacheMap {
hi := c.f.lightHouse.Query(vpnIp)
if hi == nil {
return nil
}
return hi.CopyCache()
}
// GetHostInfoByVpnAddr returns a single tunnels hostInfo, or nil if not found
// Caller should take care to Unmap() any 4in6 addresses prior to calling.
func (c *Control) GetHostInfoByVpnAddr(vpnAddr netip.Addr, pending bool) *ControlHostInfo {
var hl controlHostLister
if pending {
hl = c.f.handshakeManager
} else {
hl = c.f.hostMap
}
h := hl.QueryVpnAddr(vpnAddr)
if h == nil {
return nil
}
ch := copyHostInfo(h, c.f.hostMap.GetPreferredRanges())
return &ch
}
// SetRemoteForTunnel forces a tunnel to use a specific remote
// Caller should take care to Unmap() any 4in6 addresses prior to calling.
func (c *Control) SetRemoteForTunnel(vpnIp netip.Addr, addr netip.AddrPort) *ControlHostInfo {
hostInfo := c.f.hostMap.QueryVpnAddr(vpnIp)
if hostInfo == nil {
return nil
}
hostInfo.SetRemote(addr)
ch := copyHostInfo(hostInfo, c.f.hostMap.GetPreferredRanges())
return &ch
}
// CloseTunnel closes a fully established tunnel. If localOnly is false it will notify the remote end as well.
// Caller should take care to Unmap() any 4in6 addresses prior to calling.
func (c *Control) CloseTunnel(vpnIp netip.Addr, localOnly bool) bool {
hostInfo := c.f.hostMap.QueryVpnAddr(vpnIp)
if hostInfo == nil {
return false
}
if !localOnly {
c.f.send(
header.CloseTunnel,
0,
hostInfo.ConnectionState,
hostInfo,
[]byte{},
make([]byte, 12, 12),
make([]byte, mtu),
)
}
c.f.closeTunnel(hostInfo)
return true
}
// CloseAllTunnels is just like CloseTunnel except it goes through and shuts them all down, optionally you can avoid shutting down lighthouse tunnels
// the int returned is a count of tunnels closed
func (c *Control) CloseAllTunnels(excludeLighthouses bool) (closed int) {
shutdown := func(h *HostInfo) {
if excludeLighthouses && c.f.lightHouse.IsAnyLighthouseAddr(h.vpnAddrs) {
return
}
c.f.send(header.CloseTunnel, 0, h.ConnectionState, h, []byte{}, make([]byte, 12, 12), make([]byte, mtu))
c.f.closeTunnel(h)
c.l.Debug("Sending close tunnel message",
"vpnAddrs", h.vpnAddrs,
"udpAddr", h.GetRemote(),
)
closed++
}
// Learn which hosts are being used as relays, so we can shut them down last.
relayingHosts := map[netip.Addr]*HostInfo{}
// Grab the hostMap lock to access the Relays map
c.f.hostMap.Lock()
for _, relayingHost := range c.f.hostMap.Relays {
relayingHosts[relayingHost.vpnAddrs[0]] = relayingHost
}
c.f.hostMap.Unlock()
hostInfos := []*HostInfo{}
// Grab the hostMap lock to access the Hosts map
c.f.hostMap.Lock()
for _, relayHost := range c.f.hostMap.Indexes {
if _, ok := relayingHosts[relayHost.vpnAddrs[0]]; !ok {
hostInfos = append(hostInfos, relayHost)
}
}
c.f.hostMap.Unlock()
for _, h := range hostInfos {
shutdown(h)
}
for _, h := range relayingHosts {
shutdown(h)
}
return
}
func (c *Control) Device() overlay.Device {
return c.f.inside
}
func copyHostInfo(h *HostInfo, preferredRanges []netip.Prefix) ControlHostInfo {
chi := ControlHostInfo{
VpnAddrs: make([]netip.Addr, len(h.vpnAddrs)),
LocalIndex: h.localIndexId,
RemoteIndex: h.remoteIndexId,
RemoteAddrs: h.remotes.CopyAddrs(preferredRanges),
CurrentRelaysToMe: h.relayState.CopyRelayIps(),
CurrentRelaysThroughMe: h.relayState.CopyRelayForIps(),
CurrentRemote: h.GetRemote(),
}
for i, a := range h.vpnAddrs {
chi.VpnAddrs[i] = a
}
if h.ConnectionState != nil {
chi.MessageCounter = h.ConnectionState.messageCounter.Load()
}
if c := h.GetCert(); c != nil {
chi.Cert = c.Certificate.Copy()
}
return chi
}
func listHostMapHosts(hl controlHostLister) []ControlHostInfo {
hosts := make([]ControlHostInfo, 0)
pr := hl.GetPreferredRanges()
hl.ForEachVpnAddr(func(hostinfo *HostInfo) {
hosts = append(hosts, copyHostInfo(hostinfo, pr))
})
return hosts
}
func listHostMapIndexes(hl controlHostLister) []ControlHostInfo {
hosts := make([]ControlHostInfo, 0)
pr := hl.GetPreferredRanges()
hl.ForEachIndex(func(hostinfo *HostInfo) {
hosts = append(hosts, copyHostInfo(hostinfo, pr))
})
return hosts
}